Deep Learning-Based Framework for Automated Segmentation of Tumors in Medical Images

Automated segmentation of tumor images has become more necessary due to the growing accessibility and use of contemporary clinical imaging. However, the features needed for a simple design of medical imaging segmentation (MIS) processes are not available in the present IS systems. By accurately designating areas of concern and minimizing laborious tasks, deep learning (DL)-based automated MIS can enhance medical efficiency. Convolutional neural networks (CNNs) have gained popularity in MIS uses throughout the last ten years. Nevertheless, because of the proximity of the convolution tiers, CNNs are not exceptionally effective at acquiring long-term geographical connections. This research provides a new structure that combines two distinct algorithms, CNNs and transformers, to reliably and effectively divide the gross tumor size (GTS) in computerized tomography (CT) scans of lung cancer (LC) cases. In this system, various quality scans were employed with multi-depth bases to preserve the advantages of higher- and lower-quality scans in the DL system. Also, a deformed transformer was used to determine the long-term dependability of the features gathered. This investigation assessed the efficiency of the suggested model on an LC database containing 564 training and 114 testing scans. The DL method was compared to five distinct DL algorithms. The clinical outcomes show that the suggested structure exceeds other CNN-oriented, transformer, and composite techniques on the basis of Dice value (0.93) and Hausdorff distance (1.34). As a result, the suggested framework may enhance the performance of the automated MIS of the initial-phase LC throughout the medical process. This kind of system may be useful for real-time adaptable chemotherapy, where an effective automated MIS process is needed. The DL system, which is built on CNN and transformers, conducts automated MIS rapidly and has the possibility to improve the medical radiation process.

Ҳали таржима қилинмаган